Abstract:

The paper investigates the usage aspects of ChatGPT (version 3.5) in STEM and computing-related education processes - more precisely, we concentrate on an elective course offered to our students - Introduction to Web Programming. In this course, our students learn the basics of client/front-side web programming (using HTML5, CSS3, and plain JavaScript), as well as some elementary server-side development using PHP. With years of experience, we have collected enough student exam results to conclude that some 10% of our students are able to solve standard programming tests with high grades and provide more-less complete web pages and simple web applications. The tests are held in a controlled environment where students have to write their programs on plain paper (as much outdated and strange it may seem), with no access to a computer or the Internet.

During the last academic year, ChatGPT became quite popular and there were many headlines generating widespread attention from the media and experts across various fields. Being driven by that "madness", we started "talking" with ChatGPT about the topics of the course, as well as trying to solve some practical tasks. Even though the course covers the basics of well-known standards, it came as a surprise for the authors that ChatGPT generated quite useful answers and, even more surprisingly, great and complete solutions for many of our standard programming tests! Not only that it created valid and correct web pages and web applications, but it was able to adapt to our comments and provide the solutions we expected from our students. The paper describes some topics we "checked" with ChatGPT and the last lecture where we suggested the usage of ChatGPT to our students, as a useful learning supplement. The negative aspects of using ChatGPT (from teachers’ point of view) were also commented on, where it was advised to keep a controlled exam environment offline, with no student access to ChatGPT.

In the results section, we comment on different test requests (similar to our lecture conversations and standard exam tasks) and provide a detailed overview of ChatGPT results, including notes on initial solutions and results obtained after some adaptation. The results cover two use cases – using ChatGPT as a teaching assistant and as a code generator and program task solver. The ChatGPT results for test tasks were graded using the same assessment method as with our students (the source code analysed and checked manually) - in 20-30 seconds it solved each task, passed the exam with the highest grade, outperforming most of our students!

Keywords:

Web programming, education, ChatGPT, GPT-3.5 architecture, HTML, JavaScript, PHP.